U of M wordmark

navigation edge

 

website logo

 

 

Change Detection

To classify or not to classify

There are various ways of approaching the use of satellite imagery for determining change in urban environments. We can divide the methods for change detection into pre-classification and post-classification techniques. The pre-classification techniques apply various algorithms directly to multiple dates of satellite imagery to generate ‘‘change'' vs. ‘‘no-change'' maps. These techniques locate where changes took place but do not provide information on the nature of the change. We do not utilize pre-classification techniques for the material found on this website.

Post-classification comparison methods use separate classifications of images acquired at different times to produce difference maps. ‘‘From–to'' change information can be generated telling us how much change occurred or what areas changed from ___ to ___. The classification of each date of imagery builds a historical series that can be more easily updated and used for other applications. The post-classification comparison approach also compensates for variation in atmospheric conditions and vegetation between dates since each classification is produced independently.

Map Algebra

Whether pre- or post classification techniques are utilized, the digital nature of satellite imagery and its layout in “grids” enables us to literally add, subtract, multiply, divide or perform any mathematical formula on the aligned pixels between images. These operations are commonly referred to as “map algebra.”

When looking at change between land cover classes, the change detection process is straightforward. A pixel or grid cell will change from one land cover type to another implying change, or remain the same, implying no change. We can monitor and analyze changes between the classes, such as agricultural land transitioning to urban or forested land transitioning to agriculture. Seven classes produce a possibility of 49 changes classes, 8 classes create 64 change classes, and so on. The set of change maps displayed for the TCMA simply monitor all rural classes (agriculture, forest, wetland) that transitioned to the urban class during the time period. For more detailed analyses, the set of TCMA coverages can be downloaded and manipulated using GIS software. Corresponding statistics can also be generated, producing a census of land cover changes.

Land Cover Class

1986

1991

1998

2002

Relative Change, 1986 – 2002 (%)

Area

(000 ha)

%

Area

(000 ha)

%

Area

(000 ha)

%

Area

(000 ha)

%

Agriculture

365

47.5

342

44.4

316

41.1

310

40.3

-15.1

Urban

183

23.8

202

26.3

235

30.6

252

32.8

37.7

Forest

112

14.6

111

14.4

106

13.8

103

13.4

-8.0

Wetland

58

7.5

60

7.8

56

7.3

51

6.6

-12.1

Water

42

5.5

46

6.0

46

5.9

43

5.6

2.4

Grass

7.2

0.9

6.4

0.8

7.7

1.0

6.7

0.9

-6.9

Extraction

1.9

0.2

2.4

0.3

2.7

0.4

2.6

0.3

36.8

Summary of Landsat classification area statistics in the TCMA for 1986, 1991, 1998, and 2002

Shades of Difference

When looking at change in impervious surface we are dealing with a gradation of classes, ranging from zero to 100%. In theory we have 10,000 possible change classes, which can quickly become overwhelming. Since our research is mainly concerned with the location and intensity of impervious surface changes, it makes sense to group the change into ranges of intensity. Therefore, our impervious surface change maps reflect areas where imperviousness increased by various sets of percentages. This gives us a sense for the intensity of development within the time period viewed. For example, if an area increased in imperviousness in the 76-100% range we know that the development was quick and completely covered. On the other hand, an area that has increased by 10 - 25% can highlight areas where lower intensity development or incremental development has taken place.

Change Statistics

We can also sum and subset the total change by geographical boundaries such as counties, cities, watersheds and lakesheds. See our more detailed maps for this array of information. Change statistics allow us to compare growth patterns between years and areas. For example, the table below lists the amount of impervious surface area for several selected cities and the entire state. Between 1990 and 2000 the amount of impervious area for the entire state increased 145,830 hectares, from 1.2 to 1.9% of the total land area, an increase of 53%.

Impervious area statistics for selected cities and the state of Minnesota for 1990 and 2000

 

Total Area (ha)

1990 ISA (ha)

2000 ISA (ha)

Change (ha)

1990 % ISA

2000 % ISA

Pct. Change

St Cloud

10,405

2,066

2,894

829

19.9

27.8

40.1

Rochester

11,932

2,405

2,953

548

20.2

24.8

22.8

Bemidji

3,448

625

698

73

18.1

20.2

11.6

Brainerd

2,936

518

575

57

17.6

19.6

10.9

Fergus Falls

3,879

405

634

229

10.4

16.3

56.7

Elk River

11,344

795

1,295

500

7.0

11.4

62.9

Sauk Rapids

1,409

323

499

176

22.9

35.4

54.7

Duluth

22,601

3,055

3,044

-11

13.5

13.5

-0.4

Mankato

4,290

939

1,414

474

21.9

32.9

50.5

Owatonna

3,436

757

990

233

22.0

28.8

30.7

Eagan

8,652

2,212

2,488

276

25.6

28.8

12.5

Plymouth

9,142

1,709

2,170

461

18.7

23.7

27.0

Woodbury

9,216

1,041

1,594

552

11.3

17.3

53.1

State

21,851,634

272,863

418,693

145,830

1.2

1.9

53.4


Classification : the process of organizing satellite imagery pixels into pre-determined groups or classes based on multispectral signature/response.

Map Algebra
: performing mathematical formulas on aligned pixels between images.

Monitoring Change in the Twin Cities Metropolitan Area.
Rural land cover (agriculture, forest and wetland) that was converted to urban from 1986 to 1991, from 1991 to 1998, and from 1998 to 2002 are highlighted in green, red and yellow, respectively.

tcma